Related Experiment Video
Updated: Mar 22, 2026
05:55
Modeling the Functional Network for Spatial Navigation in the Human Brain
Published on: October 13, 2023
1.6K
Task-specific feature extraction and classification of fMRI volumes using a deep neural network initialized with a
Hojin Jang1, Sergey M Plis2, Vince D Calhoun3
1Department of Brain and Cognitive Engineering, Korea University, Seoul, Republic of Korea.
Neuroimage
|April 16, 2016
Summary
Deep neural networks (DNNs) successfully classified functional MRI volumes from sensorimotor tasks. This approach extracts task-specific brain representations, aiding potential clinical applications in neurological disease diagnosis.
Area of Science:
- Neuroscience
- Machine Learning
- Medical Imaging
Background:
- Deep neural networks (DNNs) excel in computer vision and speech processing.
- DNNs have been applied to neuroimaging data like MRI and PET.
- No prior studies utilized DNNs for 3D whole-brain fMRI volume analysis to extract task-discriminative representations.
Purpose of the Study:
- To apply fully connected feedforward DNNs to 3D whole-brain fMRI volumes.
- To extract hidden volumetric representations discriminative for specific sensorimotor tasks.
- To evaluate the DNN's classification performance and the interpretability of extracted features.
Main Methods:
- Utilized fMRI data from 12 healthy participants across four sensorimotor tasks.
- Employed a leave-one-subject-out cross-validation with a pretrained deep belief network for DNN initialization.
- Fine-tuned the DNN while controlling weight-sparsity levels to optimize fMRI volume classification.
Main Results:
- Achieved a minimum classification error rate of 6.9% (±3.8%) with a three-layer DNN under optimal sparsity.
- Outperformed single-layer (9.4%±4.6%) and two-layer (7.4%±4.1%) networks.
- DNN weights exhibited task-specific spatial patterns, and hidden layer outputs encoded task information.
Conclusions:
- DNNs can effectively classify individual fMRI volumes by extracting hidden, task-associated representations.
- The method shows promise for automatic classification and diagnosis of neuropsychiatric/neurological diseases in clinical settings.
- This approach may aid in predicting disease severity and recovery without traditional activation pattern analysis.